> 
> The initial delay of the seL4 version possibly depends on the amount of
> physical memory used. At boot time, physical memory gets converted into
> seL4 kernel objects in [1] (if I remember right). To see the effect, you
> may try tweaking the Qemu's -m argument.

not sure - delay time too big. 
Initially I think that it related to probe of big ranges of addresses used by 
attach_at() during allocator arena initialisation (I already send a letter with 
question about difference between seL4 and NOVA memory and receive answer, 
thanks). Anyway, I found that this part run relatively fast, so, not a case 
(while I need to remove error messages about failed allocation attempt).

Now start of sel4 golang code requires literally 2-3 minutes even for 512M ram 
for qemu, while for the same qemu and same code NOVA run in seconds.
I don’t able to check this with current version - seems that this is specifics 
golang runtime. Still suspect timers/etc.
Anyway, this could be a combination of timer and genode scheduler which is not 
that tailored for such workload, as I suspect.
Will continue check after finalising of porting for arm64, most of it done.

For  aarch64  it is not clear for me what to do with so called aarch64 
extensions which handled differently in libc sources [2] (has traditional 
structure ucontext for generic registers and floating point ones), and for 
glibc aarch64 [1] which have somehow similar structure for generic registers 
and just additional page attached to context where it put some magic and 
additional fields and fill by other FP registers. 
I choose to move closer to libc structure to minimise changes in related 
structures while have some doubts about different dialects of ARM64 supported. 
I run it only inside qemu.

[1] 
https://github.com/bminor/glibc/blob/master/sysdeps/unix/sysv/linux/aarch64/sys/ucontext.h
[2] 
https://github.com/freebsd/freebsd-src/blob/master/sys/arm64/include/ucontext.h


_______________________________________________
Genode users mailing list
[email protected]
https://lists.genode.org/listinfo/users

Reply via email to